A UAV path planning method and system for complex urban environments

Abstract

This invention discloses a method and system for unmanned aerial vehicle (UAV) path planning in complex urban environments, relating to the technical field of UAV path planning. This method deeply integrates multiple algorithms. Its combination, information transfer between stages, and optimization results inheritance mechanism facilitate UAV path planning. The design of an adaptive learning factor with a time-varying period significantly enhances the PSO's ability to balance global exploration and local development, enhancing population diversity and global search capabilities. Through the PSO's rapid global guidance, the SSA's enhanced population dynamics, the GA's deep solution space optimization, and the improved final accuracy of the LS algorithm, UAV paths can be planned quickly and efficiently.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicle (UAV) path planning, and in particular to a UAV path planning method and system for use in complex urban environments. Background Art

[0002] In the field of drone path planning, existing technologies have developed a variety of unique solutions, providing both theoretical and practical foundations for drone navigation in diverse environments. For example, classic graph search-based algorithms can discretize the environment into graph nodes and demonstrate excellent performance, ensuring optimal path finding. However, when faced with large-scale, high-dimensional, complex three-dimensional environments, the number of nodes and search space rapidly expand, resulting in excessive computational complexity and memory consumption, making it difficult to meet real-time requirements. The artificial potential field method (APF) has been applied in local obstacle avoidance due to its computational simplicity and fast response. It guides drone movement by constructing a virtual gravitational field (directed toward the target point) and a repulsive field (derived from obstacles). However, the APF method has an inherent drawback: it is prone to falling into local optimal solutions. This means that the drone may become trapped in potential energy depressions and unable to reach the target point, especially in environments with multiple obstacles or complex layouts (such as U-shaped traps). Rapidly expanding random trees (RRT) and its improved algorithms (such as RRT and InformedRRT) are sampling-based path planning methods that demonstrate unique advantages when dealing with high-dimensional continuous space problems. They can quickly find a feasible path and gradually explore the entire state space through random sampling. However, the paths generated by standard RRT are often not smooth enough, requiring additional smoothing, and convergence to the optimal path is relatively slow, especially in environments with narrow passages or complex constraints. Summary of the Invention

[0003] The purpose of the present invention is to provide a UAV path planning method and system for complex urban environments to improve the above technical problems.

[0004] In order to achieve the above-mentioned object of the invention, the embodiment of the present invention provides the following technical solutions:

[0005] A UAV path planning method for complex urban environments, comprising:

[0006] Obtain 3D obstacle modeling data and path node information for a city's planned scenario, and construct a corresponding 3D obstacle model.

[0007] Determine the target information of the UAV, set the planning parameters of the UAV, and construct a comprehensive obstacle avoidance cost function based on the three-dimensional obstacle model;

[0008] Based on the comprehensive obstacle avoidance cost function and the target information of each UAV, the UAV path fusion planning algorithm is used to plan the path of each UAV in the three-dimensional obstacle model to obtain the path planning results.

[0009] Furthermore, the obstacle three-dimensional modeling data includes coordinate data, plane data, elevation data, three-dimensional point cloud data, texture data, attribute data and GIS data of the obstacle;

[0010] The path node information includes the total number of path nodes and the location data of each path node;

[0011] The planning parameters include the number of drones, the drone safety cache radius, the maximum constraint altitude, the minimum constraint altitude, the turning angle threshold and the climb angle change threshold.

[0012] Furthermore, a comprehensive obstacle avoidance cost function is constructed, which specifically includes:

[0013] Determine the target information of each drone and set the planning parameters of the drone;

[0014] Obtain the real-time coordinates and real-time operating angles of each drone in real time;

[0015] Based on the path node information, the distance cost between two adjacent and continuous path nodes is calculated to construct a distance cost function;

[0016] Based on the path node information and the UAV safety cache radius, the straight-line distance between each path node and the center of different obstacles is calculated to construct the obstacle avoidance cost function;

[0017] Based on the maximum and minimum constraint altitudes of the UAV, an elevation constraint cost function is constructed;

[0018] Based on the real-time coordinates, actual operating angles, and planning parameters of each UAV, the climbing angle and turning angle of each UAV are calculated to construct a maneuver constraint cost function.

[0019] The comprehensive cost coefficient is set, and the distance cost function, obstacle avoidance cost function, elevation constraint cost function and maneuver constraint cost function are weighted and summed to construct a comprehensive obstacle avoidance cost function.

[0020] Furthermore, the process of obtaining the path planning result of any UAV includes:

[0021] Set the boundary space and the maximum number of iterations; initialize the particle swarm based on the path point information to obtain the initial particle position and initial particle velocity of each particle;

[0022] Calculate the adaptive dynamic inertia weight and adaptive learning factor of the current iteration, use the particle swarm optimization algorithm to preliminarily optimize the particle swarm, and obtain the corresponding preliminary optimized particle swarm;

[0023] The sparrow search algorithm is used to perform secondary optimization on the initial optimized particle swarm to obtain the secondary optimized particle swarm;

[0024] The genetic algorithm is used to optimize the quadratic optimized particle swarm to obtain the quadratic optimized particle swarm;

[0025] The three optimized particle swarms are optimized using the local search algorithm to obtain the final optimized particle swarm;

[0026] Based on the final optimized particle swarm, the optimal solution is determined and used as the path planning result.

[0027] Furthermore, the process of obtaining the preliminary optimized particle swarm includes:

[0028] The initial particle position of each particle is taken as the corresponding individual optimal position; the fitness of each particle is calculated using the comprehensive obstacle avoidance cost function, and the initial particle position with the highest fitness is taken as the global optimal solution;

[0029] Calculate the adaptive dynamic inertia weight based on the adaptive dynamic adjustment strategy;

[0030] Calculate the adaptive learning factor based on the dynamic adjustment mechanism; the adaptive learning factor includes individual learning factor and social learning factor;

[0031] Based on the adaptive dynamic inertia weight and adaptive learning factor of the current iteration, the initial particle velocity and initial particle position of each particle are updated;

[0032] Based on the updated velocity of each particle, the individual optimal position of each particle and the global optimal solution are updated to obtain a preliminary optimized particle swarm.

[0033] Furthermore, the acquisition process of the secondary optimization particle swarm is:

[0034] Calculate the fitness of each particle in the preliminary optimized particle swarm; based on each fitness, divide the preliminary optimized particle swarm into discoverers, followers and guards;

[0035] Use the discoverer to explore the search space, determine whether to trigger alert behavior based on the safety threshold, and update the discoverer's particle position;

[0036] Based on the particle position updated by the discoverer, the follower is used to perform local exploration and update the particle position of the follower;

[0037] Use the sentinel to perform global search and update the particle position of the sentinel;

[0038] Based on the updated particle positions corresponding to the discoverer, follower and sentinel, the corresponding fitness is calculated, the global optimal solution is updated, and the secondary optimized particle swarm is obtained.

[0039] Furthermore, the process of obtaining the particle swarm through three-step optimization includes:

[0040] The competition selection strategy is used to calculate the individual selection probability of each particle in the quadratic optimization particle swarm; the particles whose individual selection probability meets the optimal selection probability threshold range are taken as parents;

[0041] Perform binary crossover operation on the parent generation to generate new child paths by exchanging and reorganizing the paths corresponding to the parent generation to generate the offspring;

[0042] Performing adaptive polynomial mutation operation on the offspring to generate mutated offspring;

[0043] The elite survival strategy is used to merge the parent generation and the mutated offspring, calculate the fitness of each individual, and select the individuals that meet the fitness threshold as the particle swarm for the third optimization.

[0044] A UAV path planning system for complex urban environments, comprising:

[0045] The obstacle 3D model construction module is used to obtain the obstacle 3D modeling data and path node information of the scene to be planned in a certain city and construct the corresponding obstacle 3D model;

[0046] The comprehensive obstacle avoidance cost function construction module is used to determine the target information of the UAV, set the planning parameters of the UAV, and construct a comprehensive obstacle avoidance cost function based on the three-dimensional obstacle model;

[0047] The UAV path planning module is used to plan the path of each UAV in the three-dimensional obstacle model based on the comprehensive obstacle avoidance cost function and the target information of each UAV, and obtain the path planning results.

[0048] The beneficial effects of the present invention are:

[0049] This method deeply integrates multiple algorithms. Its combination mode and the information transmission and optimization result inheritance mechanism between each stage are beneficial to the path planning of UAVs. In the particle swarm optimization (PSO) stage, an adaptive learning factor with a time-varying period is designed to significantly improve the PSO's global exploration and local development balance capabilities. In the sparrow search algorithm (SSA) stage, the role allocation mechanism (including discoverers, sentinels, and followers) and their respective unique behavior strategies and position update rules are optimized to enhance population diversity and global search capabilities. Through the rapid global guidance of PSO, the population dynamics enhancement of SSA, the deep optimization of the solution space of GA, and the final accuracy improvement of LS, a multi-stage and progressive collaborative optimization framework is formed, which can efficiently and robustly solve the three-dimensional UAV path planning problem under complex constraints.

[0050] This system integrates multiple hybrid metaheuristic algorithms to construct a multi-stage collaborative optimization framework, which is conducive to improving the global exploration and local development balance capabilities of the UAV path fusion planning algorithm; the system structure is simple, and the UAV path is planned through a multi-stage optimization algorithm, which quantifies the path length, obstacle avoidance effect, flight altitude and flight maneuverability, and can plan the path quickly and efficiently. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without making any creative efforts.

[0052] Figure 1 A flow chart of a method in an embodiment of the present invention;

[0053] Figure 2 This is a system structure diagram of this embodiment;

[0054] Figure 3 This is the optimal value curve diagram of the evaluation function F1 in this embodiment;

[0055] Figure 4 This is the optimal value curve diagram of the evaluation function F2 in this embodiment;

[0056] Figure 5 This is the optimal value curve diagram of the evaluation function F3 in this embodiment;

[0057] Figure 6 This is a simulation scene result diagram of scene 1 in this embodiment;

[0058] Figure 7 This is a simulation scene result diagram of scene 2 in this embodiment;

[0059] Figure 8 This is a simulation scene result diagram of scene 3 in this embodiment;

[0060] Figure 9 This is a simulation scene result diagram of scene 4 in this embodiment. DETAILED DESCRIPTION

[0061] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, not all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present invention.

[0062] See also Figure 1 This embodiment provides a UAV path planning method for a complex urban environment, which includes:

[0063] S1. Obtain 3D obstacle modeling data and path node information for a city scenario to be planned, and construct a corresponding 3D obstacle model. The 3D obstacle modeling data includes obstacle coordinate data, plane data, elevation data, 3D point cloud data, texture data, attribute data, and GIS data. The path node information includes the total number of path nodes and the location data of each path node.

[0064] Construct a 3D model of the obstacle. Specifically, perform data cleaning, correction, fusion, and feature extraction on the 3D obstacle model data to obtain characteristic information of different obstacles (such as buildings) in the scene, such as edges, outlines, and roof shapes. 3D modeling can be performed using CAD / BIM-based modeling methods, oblique photogrammetry-based modeling methods, or LiDAR point cloud-based modeling methods. Based on this characteristic information, 3D modeling is performed on different obstacles to obtain the corresponding 3D obstacle models.

[0065] S2. Determine the target information of the drone, set the planning parameters of the drone, and construct a comprehensive obstacle avoidance cost function based on the three-dimensional obstacle model. In this embodiment, the target information is the coordinate data of the drone's destination.

[0066] The planning parameters include the number of drones, drone safety cache radius, maximum constraint altitude, minimum constraint altitude, turning angle threshold, and climb angle change threshold.

[0067] The comprehensive obstacle avoidance cost function includes distance cost function, obstacle avoidance cost function, elevation constraint cost function and maneuver constraint cost function.

[0068] Thus, the S2 includes:

[0069] S2-1. Determine the target information of each drone and set the planning parameters of the drone.

[0070] S2-2. Obtain the real-time coordinates and real-time operating angle of each UAV in real time; the real-time operating angle includes the climbing angle and the turning angle.

[0071] S2-3, based on the path node information, calculate the distance cost of two adjacent and continuous path nodes and construct a distance cost function; thus, the distance cost function The corresponding formula is:

[0072] ;

[0073] in, represents the summation function, Indicates the total number of path nodes, 、 Respectively represent Waypoints, The x-axis coordinates of the path points, 、 Respectively represent Waypoints, The y-axis coordinate of the path point, 、 Respectively represent Waypoints, The z-axis coordinate of each path point, represents the L2 norm (Euclidean norm). In this embodiment, the coordinate system is a three-dimensional Cartesian coordinate system.

[0074] S2-4, based on the path node information and the UAV safety cache radius, calculate the straight-line distance between each path node and the center of different obstacles, and construct the obstacle avoidance cost function; thus, the obstacle avoidance cost function The corresponding formula is:

[0075] ;

[0076] in, Indicates the number of obstacles, Indicates the number of path segments between two path nodes. Indicates the The radius of the obstacle, Indicates the The path segment is The straight-line distance from the center of the obstacle, 、 They represent the UAV body radius and the UAV safety cache radius respectively.

[0077] S2-5, based on the maximum constraint altitude and the minimum constraint altitude of the UAV, construct the elevation constraint cost function; thus, the elevation constraint cost function The corresponding formula is:

[0078] ;

[0079] in, represents the absolute value function, 、 Respectively represent the maximum constraint altitude and minimum constraint altitude of the UAV, Indicates the The actual altitude of the drone.

[0080] S2-6. Based on the real-time coordinates, actual operating angles, and planning parameters of each UAV, the climbing angle and turning angle of each UAV are calculated to construct a maneuver constraint cost function.

[0081] Thus, the maneuver constraint cost function The corresponding formula is:

[0082] ;

[0083] ;

[0084] ;

[0085] in, 、 Respectively represent The climbing angle and turning angle of each drone, represents the azimuth angle calculation function, Indicates the turning angle threshold allowed for the drone. represents the maximum function, 、 Respectively represent The drone in Obstacles, The climb angle at the obstacle, Indicates the threshold value of the climb angle change allowed for the drone. Indicates the The drone in The turning angle at the obstacle.

[0086] S2-7, set the comprehensive cost coefficient, perform weighted summation on the distance cost function, obstacle avoidance cost function, elevation constraint cost function and maneuver constraint cost function, and construct a comprehensive obstacle avoidance cost function. Thus, the comprehensive obstacle avoidance cost function The corresponding formula is:

[0087] ;

[0088] in, 、 、 、 Represents the comprehensive cost coefficient.

[0089] This embodiment designs a comprehensive obstacle avoidance cost function that comprehensively considers and quantifies path length, obstacle avoidance effect, satisfaction of flight altitude constraints, and compliance with flight maneuverability constraints, which is the key to achieving effective path evaluation and guidance.

[0090] S3. Based on the comprehensive obstacle avoidance cost function and the target information of each UAV, the UAV path fusion planning algorithm is used to plan the path of each UAV in the three-dimensional obstacle model to obtain the path planning results.

[0091] The UAV path fusion planning algorithm is the PSO-SSA-GA-LS algorithm, which includes the particle swarm optimization algorithm (PSO) stage, the sparrow search algorithm (SSA) stage, the genetic algorithm (GA) stage and the local search algorithm (LS) stage.

[0092] Since there are multiple drones and each drone has corresponding target information, taking any drone as an example, the drone path fusion planning algorithm includes:

[0093] S3-1. Set the boundary space and the maximum number of iterations. Initialize the particle swarm based on the path point information, that is, initialize the particle position and particle velocity, and obtain the initial particle position and initial particle velocity of each particle. Each particle in the particle swarm represents a candidate path, and each candidate path includes multiple path nodes.

[0094] S3-2. Calculate the adaptive dynamic inertia weight and adaptive learning factor of the current iteration, and use the particle swarm optimization algorithm to preliminarily optimize the particle swarm to obtain the corresponding preliminary optimized particle swarm.

[0095] The S3-2 includes:

[0096] S3-2-1. The initial particle position of each particle is used as the corresponding individual optimal position; the fitness of each particle is calculated using the comprehensive obstacle avoidance cost function, and the initial particle position with the highest fitness is used as the global optimal solution;

[0097] S3-2-2. Calculate the adaptive dynamic inertia weight based on the adaptive dynamic adjustment strategy. The corresponding formula is:

[0098] ;

[0099] ;

[0100] in, 、 Respectively represent The current iteration and The adaptive perturbation parameter of the iteration, represents the perturbed random value, 、 denote the cosine function and the inverse cosine function, respectively. Indicates the Adaptive dynamic inertia weight of iterations, 、 Represent the maximum and minimum values ​​of the adaptive dynamic inertia weight respectively.

[0101] S3-2-3. Calculate the adaptive learning factor based on the dynamic adjustment mechanism. The adaptive learning factor includes the individual learning factor and the social learning factor. The corresponding formula is:

[0102] ;

[0103] ;

[0104] in, represents the maximum number of iterations, Indicates the The individual learning factor of the iteration, 、 Represent the maximum and minimum values ​​of the individual learning factor, represents the sine function, Indicates the The social learning factor of the iteration, 、 Represent the maximum and minimum values ​​of the social learning factor respectively.

[0105] S3-2-4. Based on the adaptive dynamic inertia weight and adaptive learning factor of the current iteration, update the initial particle velocity and initial particle position of each particle. The corresponding formula is:

[0106] ;

[0107] ;

[0108] in, 、 Respectively expressed in Iteration, The first iteration The updated particle velocity of each particle and the particle velocity before the update (initial particle velocity), 、 Respectively expressed in Iteration, The first iteration The particle positions before and after the update of the particle, Indicates the The individual optimal position of each particle, represents the global optimal solution.

[0109] S3-2-5. Based on the updated velocity of each particle, update the individual optimal position of each particle and the global optimal solution to obtain a preliminary optimized particle swarm. The method for updating the individual optimal position and the global optimal solution is the existing PSO algorithm, that is, iterating according to the classic update rule of PSO.

[0110] S3-3. Use the sparrow search algorithm to perform secondary optimization on the initial optimized particle swarm to obtain a secondary optimized particle swarm.

[0111] After initial optimization of the PSO algorithm, the Sparrow Search Algorithm (SSA) was introduced to further enhance the population's diversity and global search capabilities. The SSA algorithm simulates the foraging and anti-predator behaviors of a sparrow colony. Its core innovation lies in a dynamic role assignment mechanism: based on the historically optimal fitness values ​​of particles (paths), the population is divided into "discoverers," "followers," and "sentinels." The optimal solution is achieved through the co-evolution of discoverers, followers, and sentinels. Its position update mechanism integrates bounds constraint checking and an efficient evaluation strategy: at each iteration, the newly generated position vector passes bounds constraint checking to ensure parameter validity. The system dynamically monitors the magnitude of particle displacement, reassessing the fitness values ​​only for individuals with significant position changes and updating their historical best records accordingly. When individual improvements lead to an update to the global optimal solution, a stagnation counter is synchronously reset to maintain exploration activity. The final positions of all particles are uniformly updated after role behavior simulation and constraint processing.

[0112] Therefore, the S3-3 includes:

[0113] S3-3-1. Calculate the fitness of each particle in the preliminary optimized particle swarm. Based on each fitness, divide the preliminary optimized particle swarm into discoverers, followers, and sentinels. Arrange the fitnesses from largest to smallest, with the top 20% of particles as discoverers, the bottom 10% as sentinels, and the remaining particles as followers.

[0114] S3-3-2. Use the discoverer to explore the search space, determine whether to trigger the alert behavior based on the safety threshold, and update the discoverer's particle position. The corresponding formula is:

[0115] ;

[0116] in, represents the exponential function with the natural constant e as the base, 、 Respectively represent Iteration, The first iteration The position of the particle of the first discoverer, The updated particle position of each discoverer and the particle position before the update, 、 、 They represent the decay coefficient, the finder index value, and the dynamic adjustment factor that control the exponential decay rate in the local refinement strategy, respectively. 、 、 Represent environmental risk monitoring indicators, random disturbance amplitude and random direction components respectively, Represents predefined key thresholds for environmental risk index.

[0117] S3-3-3. Based on the particle position updated by the discoverer, use the follower to perform local exploration and update the particle position of the follower. The corresponding formula is:

[0118] ;

[0119] in, represents the subset of followers with the top 50% in the follower set, represents the subset of followers in the bottom 50% of the follower set, 、 Respectively represent Iteration, The first iteration The particle position of the follower, i.e. The updated particle position of each follower and the particle position before the update, Indicates the The worst particle under iterations, 、 、 Respectively represent the follower risk factor, stability factor and follower index value, Indicates the The best particle in the iteration, Represents the sum of index values. The best particle is the particle position with the highest fitness among the positions of the discoverer before and after the update in the current iteration, and the worst particle is the particle position with the lowest fitness among the positions of the discoverer, follower, and sentinel before and after the update in the current iteration.

[0120] S3-3-4. Use the sentinel to perform a global search and update the particle position of the sentinel. The corresponding formula is:

[0121] ;

[0122] in, Indicates the index value of the sentinel. 、 Respectively represent Iteration, The first iteration The particle position of the follower, i.e. The updated particle position of each follower and the particle position before the update, represents the step size parameter of the non-elite vigilante, controlling the magnitude of the movement towards the global optimal position, Indicates the direction in which the sparrow moves and participates in controlling the step length of the random walk when the alerter realizes the danger. 、 represent non-marginal individuals and marginal individuals respectively, express The fitness value of express The fitness value of .

[0123] S3-3-5. Based on the updated particle positions corresponding to the discoverer, follower, and sentinel, calculate the corresponding fitness, update the global optimal solution, and use the particle position with the highest fitness as the updated global optimal solution; integrate the updated particle positions and the updated global optimal solution to obtain a secondary optimized particle swarm; the secondary optimized particle swarm includes the updated positions corresponding to the discoverer, follower, and sentinel and the updated global optimal solution.

[0124] This embodiment introduces a sparrow search algorithm, whose role allocation and behavior strategy form a closed-loop collaborative search mechanism, effectively enhancing the algorithm's ability to escape from local optimality.

[0125] S3-4. Use genetic algorithm to optimize the secondary optimized particle swarm to obtain the tertiary optimized particle swarm.

[0126] The S3-4 includes:

[0127] S3-4-1. Calculate the individual selection probability of each particle in the quadratic optimization particle swarm using the competition selection strategy , the corresponding formula is:

[0128] ;

[0129] in, represents the intensity parameter, which is used to control the intensity of the selection pressure. represents a random number generation function, represents the total number of particles in the quadratic optimization particle swarm, Represents the set of individuals currently participating in particle selection, that is, the secondary optimization particle swarm.

[0130] S3-4-2. The particle whose individual selection probability reaches the optimal selection probability threshold is selected as the parent generation;

[0131] S3-4-3. Perform a binary crossover operation on the parent generation. By exchanging and recombining certain characteristic fragments of the parent generation's path (such as path point coordinates), a new child generation path is generated. This integrates the characteristics of the excellent parent generation and explores new solution combinations to generate the child generation. The formula corresponding to the binary crossover operation is:

[0132] ;

[0133] in, Represents offspring, represents the location of sub-solution generation in the control crossover operation, Modeling the distribution exponential representing the binary crossover to control the distribution range of the offspring solutions.

[0134] S3-4-3, perform an adaptive polynomial mutation operation on the offspring to generate the mutated offspring. The formula corresponding to the adaptive polynomial mutation operation is:

[0135] ;

[0136] in, represents the offspring after mutation, Indicates the adjustment amplitude of the offspring solution in the mutation operation, The distribution index of the polynomial variance controls the distribution characteristics of the disturbance, Represents the maximum permissible perturbation range to ensure that the variational results meet the boundary constraints.

[0137] This example introduces a genetic algorithm to further optimize and evolve the population optimized by PSO and SSA. GA simulates operations such as selection, crossover, and mutation in biological evolution, performing small random perturbations on certain path parameters to introduce new genetic diversity, prevent premature population convergence, and enhance the algorithm's local search capabilities. After generating the offspring population, an elite survival strategy is employed to merge the parent and offspring populations and select the optimal individuals based on fitness to advance to the next generation, ensuring the stable inheritance of excellent solutions.

[0138] S3-4-4. Use the elite survival strategy to merge the parent generation and the mutated offspring, calculate the fitness of each individual, select the individuals that meet the fitness threshold and use them as the particle swarm for the third optimization. Set the optimal selection probability threshold and fitness threshold according to actual needs.

[0139] S3-5. Use the local search algorithm to optimize the three-time optimized particle swarm to obtain the final optimized particle swarm.

[0140] The local search algorithm adopts the Pattern Search search strategy, and the corresponding formula is:

[0141] ;

[0142] in, represents the potential refined solution after local search, represents the global minimum function, 、 Represents the first particles The upper and lower boundaries, i.e. the boundary space, are used to prevent particles from crossing the boundary. Indicates that the mode The cost or performance metric function under .

[0143] This embodiment adopts a Pattern Search search strategy, which can be intelligently triggered when the current iteration shows signs of stagnation, thereby performing refined local development of the currently found optimal solution or adjacent areas to find a more accurate path solution without significantly increasing the overall computational complexity.

[0144] S3-6. Based on the final optimized particle swarm, determine the optimal solution and use the optimal solution as the path planning result.

[0145] S3-6 includes:

[0146] S3-6-1. Calculate the fitness of each particle in the final optimized particle swarm, and take the position of the particle with the highest fitness as the initial optimal solution.

[0147] S3-6-2. Determine whether the number of current iterations reaches the number of iterations; if so, take the initial optimal solution as the optimal solution, that is, the path planning result; otherwise, add 1 to the number of current iterations and return to S3-1. The initial value of the number of current iterations is 0.

[0148] Repeat steps S3-1 to S3-6 until the path planning results of all drones are obtained.

[0149] This embodiment connects and deeply integrates multiple algorithms in series. Its combination method and the information transmission and optimization result inheritance mechanism between each stage are beneficial to the path planning of UAVs. In the particle swarm optimization (PSO) stage, an adaptive learning factor with a time-varying period is designed to significantly improve the PSO's global exploration and local development balance capabilities. In the sparrow search algorithm (SSA) stage, the role allocation mechanism (including discoverers, sentinels, and followers) and their respective unique behavior strategies and position update rules are optimized to enhance population diversity and global search capabilities. Through the rapid global guidance of PSO, the population dynamics enhancement of SSA, the deep optimization of the solution space of GA, and the final accuracy improvement of LS, a multi-stage and progressive collaborative optimization framework is formed, which can efficiently and robustly solve the three-dimensional UAV path planning problem under complex constraints.

[0150] like Figure 2 As shown, a UAV path planning system for complex urban environments includes:

[0151] The obstacle 3D model construction module is used to obtain the obstacle 3D modeling data and path node information of the scene to be planned in a certain city and construct the corresponding obstacle 3D model;

[0152] The comprehensive obstacle avoidance cost function construction module is used to determine the target information of the UAV, set the planning parameters of the UAV, and construct a comprehensive obstacle avoidance cost function based on the three-dimensional obstacle model;

[0153] The UAV path planning module is used to plan the path of each UAV in the three-dimensional obstacle model based on the comprehensive obstacle avoidance cost function and the target information of each UAV, and obtain the path planning results.

[0154] This system integrates multiple hybrid metaheuristic algorithms to construct a multi-stage collaborative optimization framework, which is conducive to improving the global exploration and local development balance capabilities of the UAV path fusion planning algorithm; the system structure is simple, and the UAV path is planned through a multi-stage optimization algorithm, which quantifies the path length, obstacle avoidance effect, flight altitude and flight maneuverability, and can plan the path quickly and efficiently.

[0155] In this embodiment, the classical particle swarm optimization algorithm (PSO), Harris Hawk Optimizer (HHO), standard sparrow search algorithm (SSA), African vulture optimization algorithm (AVOA) and spotted hyena optimizer (SHO) are used for comparison with the method of the present invention.

[0156] The population size was set to 50 particles or individuals, the maximum number of iterations was 500, and the internationally accepted CEC2020 standard test function set was used as a benchmark for evaluating algorithm performance. This test set contains 10 optimization problems (F1-F10) with varying complexity (e.g., unimodal, multimodal, mixed, combined, separable, inseparable, rotated, and offset).

[0157] PSO, HHO, SSA, AVOA, SHO and this method (PG-SSA) are used to plan the path of the UAV in the same scenario.

[0158] Table 1

[0159]

[0160] As shown in Table 1, in the evaluation function F1, Figure 3 As shown in Figure 2, the optimal value, average value, and standard deviation of the present invention are 1.00e+02, 1.75e+02, and 2.18e+02, respectively, which are all smaller than the optimal values, average values, and standard deviations corresponding to PSO, HHO, SSA, AVOA, and SHO. In the evaluation function F2, as Figure 4 As shown in Figure 2, the optimal value, average value, and standard deviation of the present invention are 1.79e+03, 2.52e+03, and 4.23e+02, respectively, which are all smaller than the optimal values, average values, and standard deviations corresponding to PSO, HHO, SSA, AVOA, and SHO. In the evaluation function F3, as Figure 5 As shown, the optimal value, average value, and standard deviation of the present invention are 7.61e+02, 7.86e+02, and 1.30e+01, respectively, which are all smaller than the optimal values, average values, and standard deviations of PSO, HHO, SSA, AVOA, and SHO. In evaluation functions F1 to F10, the figure of merit, average value, and standard deviation of the present invention are all smaller than the optimal values, average values, and standard deviations of PSO, HHO, SSA, AVOA, and SHO.

[0161] On evaluation function F1, the average value of PG-SSA is higher than that of the PSO algorithm and far outperforms the other compared algorithms. On evaluation function F8, the average value of PG-SSA is 8.3% lower than that of AVOA, further demonstrating the high optimization accuracy of the PG-SSA algorithm.

[0162] PG-SSA achieved the lowest standard deviation index on all eight evaluation functions from F2 to F9, indicating that the convergence process of PG-SSA is highly consistent, the volatility of the results of multiple runs is small, the robustness is strong, and the convergence stability is better than other optimization algorithms.

[0163] Depend on Figures 3 to 5 It can be seen that PG-SSA shows faster convergence speed and stronger ability to escape local optimality on the test function, and can continuously evolve towards a better solution.

[0164] The comprehensive cost function design and constraint handling mechanisms at each stage of the algorithm enable PG-SSA to effectively find high-quality feasible paths under these complex constraints. Its stable performance across various threat levels and simulated real-world scenarios, as well as achieving optimal standard deviations (≤0.03) on 80% of functions in the CEC2020 benchmark, demonstrate its excellent convergence stability and adaptability to diverse and complex environments.

[0165] In another embodiment, drone path planning was performed for four scenarios using PSO, AVOA, HHO, SSA, GTO (Gorilla Troop Optimization), and the present method (PG-SSA). Each method used the same input data, model constraints, and other parameters. The 3D obstacle models for each scenario were 3D geometric models.

[0166] The four scenarios are Scenario 1 (basic threat), Scenario 2 (medium complexity), Scenario 3 (high complexity) and Scenario 4 (real environment simulation).

[0167] Table 2

[0168]

[0169] like Figure 6 As shown in Table 2, in scenario 1, the trajectory length generated by PG-SSA is 702.53 meters, which is shorter than the trajectory lengths generated by GTO (788.86 meters) and PSO (789.03 meters), respectively, with improvements of 10.9% and 11%, and shorter than the trajectory lengths generated by SSA (794.99 meters), AVOA (814.80 meters), and HHO (813.19 meters), respectively, with improvements of 11.6%, 13.7%, and 15.7%.

[0170] like Figure 7As shown in Table 2, in scenario 2, the trajectory length generated by PG-SSA is 541.55 meters, which is shorter than the trajectory lengths generated by SSA (561.87 meters), PSO (620.96 meters), AVOA (671.01 meters), HHO (641.65 meters), and GTO (638.98 meters), with improvements of 3.6%, 14.1%, 20.8%, 14.9%, and 15.2%, respectively.

[0171] like Figure 8 As shown in Table 2, in scenario 3, the trajectory length generated by PG-SSA is 1652.45 meters, which is optimized by 9.8%, 22.9%, 6.9%, 10.5% and 17.9% compared with the trajectory lengths generated by PSO (1832.27 units), AVOA (2144.65 meters), HHO (1776.84 meters), GTO (1846.77 meters) and SSA (2014.57 meters), respectively, demonstrating its strong optimization ability in complex constrained environments.

[0172] like Figure 9 As shown in Table 2, in scenario 4, the trajectory length generated by G-SSA is 518.98 meters, which is better than GTO (530.71 meters), PSO (601.36 meters), AVOA (657.46 meters), HHO (581.79 meters), and SSA (566.59 meters), with improvements of 2.2%, 13.6%, 21%, 10.7%, and 8.4%, respectively.

[0173] In summary, PG-SSA can stably find high-quality flight paths in environments of different complexities. Even in scenarios with dense obstacles and strict constraints, PG-SSA can still effectively perform global search and local optimization to plan a safe and economical path, further fully proving that the present invention has significant performance advantages in actual application scenarios. It is superior to or significantly superior to the existing mainstream algorithms compared in terms of optimization accuracy, convergence stability, environmental adaptability, and the ability to solve complex multi-constraint optimization problems, providing an innovative and effective technical solution for efficient and safe three-dimensional path planning of drones in complex urban environments. The above is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed by the present invention, which should be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A UAV path planning method for complex urban environments, characterized in that: include: Obtain 3D obstacle modeling data and path node information for a city's planned scenario, and construct a corresponding 3D obstacle model. Determine the target information of the UAV, set the planning parameters of the UAV, and construct a comprehensive obstacle avoidance cost function based on the three-dimensional obstacle model; Based on the comprehensive obstacle avoidance cost function and the target information of each UAV, the UAV path fusion planning algorithm is used to plan the path of each UAV in the three-dimensional obstacle model to obtain the path planning results; Construct a comprehensive obstacle avoidance cost function, specifically including: Determine the target information of each drone and set the planning parameters of the drone; Obtain the real-time coordinates and real-time operating angles of each drone in real time; Based on the path node information, the distance cost between two adjacent and continuous path nodes is calculated to construct a distance cost function; Based on the path node information and the UAV safety cache radius, the straight-line distance between each path node and the center of different obstacles is calculated to construct the obstacle avoidance cost function; the obstacle avoidance cost function The corresponding formula is: ； in, represents the summation function, Indicates the total number of path nodes, Indicates the number of obstacles, Indicates the number of path segments between two path nodes. Indicates the The radius of the obstacle, Indicates the The path segment is The straight-line distance from the center of the obstacle, 、 They represent the UAV body radius and the UAV safety cache radius respectively; Based on the maximum and minimum constraint altitudes of the UAV, an elevation constraint cost function is constructed; Based on the real-time coordinates, actual operating angles and planning parameters of each UAV, the climbing angle and turning angle of each UAV are calculated to construct the maneuver constraint cost function; the maneuver constraint cost function The corresponding formula is: ； ； ； in, 、 Respectively represent The climbing angle and turning angle of each drone, represents the azimuth angle calculation function, Indicates the turning angle threshold allowed for the drone. represents the maximum function, 、 Respectively represent The drone in Obstacles, The climb angle at the obstacle, Indicates the threshold value of the climb angle change allowed for the drone. Indicates the The drone in The turning angle at the obstacle, 、 Respectively represent Waypoints, The x-axis coordinates of the path points, 、 Respectively represent Waypoints, The y-axis coordinate of the path point, 、 Respectively represent Waypoints, The z-axis coordinate of each path point, represents the absolute value function; Set the comprehensive cost coefficient, perform weighted summation of the distance cost function, obstacle avoidance cost function, elevation constraint cost function, and maneuver constraint cost function to construct a comprehensive obstacle avoidance cost function; The process of obtaining the path planning results of any UAV includes: Set the boundary space and the maximum number of iterations; initialize the particle swarm based on the path node information to obtain the initial particle position and initial particle velocity of each particle; Calculate the adaptive dynamic inertia weight and adaptive learning factor of the current iteration, use the particle swarm optimization algorithm to preliminarily optimize the particle swarm, and obtain the corresponding preliminary optimized particle swarm; The sparrow search algorithm is used to perform secondary optimization on the initial optimized particle swarm to obtain the secondary optimized particle swarm; The genetic algorithm is used to optimize the quadratic optimized particle swarm to obtain the quadratic optimized particle swarm; The three optimized particle swarms are optimized using the local search algorithm to obtain the final optimized particle swarm; Based on the final optimized particle swarm, the optimal solution is determined and used as the path planning result.

2. The method for UAV path planning in a complex urban environment according to claim 1, characterized in that: The obstacle three-dimensional modeling data includes coordinate data, plane data, elevation data, three-dimensional point cloud data, texture data, attribute data and GIS data of the obstacle; The path node information includes the total number of path nodes and the location data of each path node; The planning parameters include the number of drones, the drone safety cache radius, the maximum constraint altitude, the minimum constraint altitude, the turning angle threshold and the climb angle change threshold.

3. The method for UAV path planning in a complex urban environment according to claim 1, characterized in that: The process of obtaining the preliminary optimized particle swarm includes: The initial particle position of each particle is taken as the corresponding individual optimal position; the fitness of each particle is calculated using the comprehensive obstacle avoidance cost function, and the initial particle position with the highest fitness is taken as the global optimal solution; Calculate the adaptive dynamic inertia weight based on the adaptive dynamic adjustment strategy; Calculate the adaptive learning factor based on the dynamic adjustment mechanism; the adaptive learning factor includes individual learning factor and social learning factor; Based on the adaptive dynamic inertia weight and adaptive learning factor of the current iteration, the initial particle velocity and initial particle position of each particle are updated; Based on the updated velocity of each particle, the individual optimal position of each particle and the global optimal solution are updated to obtain a preliminary optimized particle swarm.

4. The method for UAV path planning in a complex urban environment according to claim 1, characterized in that: The acquisition process of the secondary optimization particle swarm is: Calculate the fitness of each particle in the preliminary optimized particle swarm; based on each fitness, divide the preliminary optimized particle swarm into discoverers, followers and guards; Use the discoverer to explore the search space, determine whether to trigger alert behavior based on the safety threshold, and update the discoverer's particle position; Based on the particle position updated by the discoverer, the follower is used to perform local exploration and update the particle position of the follower; Use the sentinel to perform global search and update the particle position of the sentinel; Based on the updated particle positions corresponding to the discoverer, follower and sentinel, the corresponding fitness is calculated, the global optimal solution is updated, and the secondary optimized particle swarm is obtained.

5. The method for UAV path planning in a complex urban environment according to claim 1, characterized in that: The acquisition process of the three-time optimized particle swarm includes: The competition selection strategy is used to calculate the individual selection probability of each particle in the quadratic optimization particle swarm; the particle whose probability reaches the optimal selection probability threshold is selected as the parent; Perform binary crossover operation on the parent generation to generate new child paths by exchanging and reorganizing the paths corresponding to the parent generation to generate the offspring; Performing adaptive polynomial mutation operation on the offspring to generate mutated offspring; The elite survival strategy is used to merge the parent generation and the mutated offspring, calculate the fitness of each individual, and select the individuals that meet the fitness threshold as the third-order optimized particle swarm.

6. A UAV path planning system for complex urban environments, used to implement the UAV path planning method for complex urban environments according to any one of claims 1 to 5, characterized in that: include: The obstacle 3D model construction module is used to obtain the obstacle 3D modeling data and path node information of the scene to be planned in a certain city and construct the corresponding obstacle 3D model; The comprehensive obstacle avoidance cost function construction module is used to determine the target information of the UAV, set the planning parameters of the UAV, and construct a comprehensive obstacle avoidance cost function based on the three-dimensional obstacle model; The UAV path planning module is used to plan the path of each UAV in the three-dimensional obstacle model based on the comprehensive obstacle avoidance cost function and the target information of each UAV, and obtain the path planning results.

Images